• 500,000 participants
• 100,000 inlucded in the sub-study
• Can infer movement with longitudinal follow-up

• Tri-axial Axtivity 100Hz over 7 days
• Started at 10AM and ended at 10PM
• Data measured in milli-g (1\(g\) = 9.80665 \(\frac{m}{s^2}\))
  • not counts or steps as other devices

Many people DIED before being able to be asked

• Data at varyling levels
  • Axtivity CWA format (Highest resolution, 100Hz)
    • very large for 100K subjects (Estimate size)
• 5 second level data
  • imputation/processing done
  • averaged into minute-level data
• Overall statistics (mean/median): overall, daily, hourly, day of week
  • removed “non-wear” periods

• Data at varyling levels
  • Axtivity CWA format (Highest resolution, 100Hz)
    • very large for 100K subjects
• 5 second level data
  • imputation/processing done
  • averaged into minute-level data
• Overall statistics (mean/median): overall, daily, hourly, day of week
  • removed “non-wear” periods

Auto-calibration [@van2014autocalibration]

• 10\(s\) window all axes SD \(< 13.0\) m\(g\).
• fit a unit gravity sphere using OLS.
• If 3 axes had values outside a \(\pm 300\)m\(g\) range - use calibration coefficient from that person
• if not use next peron’s accelerometer record from the same device

• recording errors and ‘interrupts’ flagged (plug in accelerometer)
• \(\pm8g\) flagged
• Resampled to 100 Hz (interrupts > 5 seconds set to missing)
• Euclidean norm, fourth order Butterworth low pass filter (f = 20Hz).
• Subtract \(1g\), negative values set to \(0\)

@doherty2017large

• Explore the data
• Get similar findings to some [@doherty2017large] analysis
• Assess “bias” in different devices (see if autocalibration is working)

• Average all data - regardless of mulitple visits per person

• Average all data - regardless of mulitple visits per person

1. Run full - show weird shit
2. Run only 95% data - show how some goes away

• say maybe solved

1. Run by day - show all 7 and the weird things
2. Show how threshold throws out day 0 and 7 (probably good thing) - peak problem
3. Show the jumps still
4. Well maybe it’s the mean - just use the median - should fix it - nope
5. Show 1-2 raw data (Need this)

• Only keep when 95% (1368 minutes) are not missing

• Take Day 0 - 7, day 0 is when they started wearing it
  • this is what we saw in the data though it should be 0-6
• Calculate statistic over all participants
  • the same minute across people
  • each day separately

“We removed non-wear time, defined as consecutive stationary episodes lasting for at least 60 minutes where all three axes had a standard deviation of less than 13.0 m\(g\)”

@doherty2017large

“Here, 13 m\(g\) was selected just above the empirically derived baseline (noise) standard deviation of 10 m\(g\) to retain only nonmovement periods.” @van2014autocalibration

1. Start off smaller than 100K people
2. Inspect the raw(ish) data
3. Autocalibration seems to work well on gross features (with 100K people)
4. Artifacts still seem present in the data
5. Back to the 100Hz data we go!

“We invited some participants to wear an activity monitor for a week, four times a year. … finished in early 2019.”